2026年7月10-12日，我院举办第二届算法、机器学习、图像处理国际学术会议，特邀7位国内外相关领域专家学者作学术报告，欢迎自动化学院及全校相关教师、博士生、硕士生参加！
报告地点：新主楼五楼报告厅、E1404
报告时间：2026年7月11日，8:30-17:00
报告专家：Qinglong Han, Swinburne University of Technology, Australia
报告题目：Driving the Futureof Transport: Communication-Efficient and Cyber-Secure Coordination Control
报告摘要：The evolution of intelligent transportation systems hinges on the seamless integration of connected and automated technologies, yet challenges in communication efficiency and cybersecurity remain critical barriers to their widespread adoption. This Keynote Address explores innovative solutions to drive the future of transport, focusing on two pivotal research domains: communication-efficient coordination control and cyber-secure coordination control for connected automated vehicle (CAV) platoons and connected railway systems. The Keynote Address begins with a concise overview of connected automated transport, followed by an in-depth exploration of key design and implementation challenges in such systems. Novel event-triggered coordination control strategies that dynamically schedule vehicle-to-vehicle and train-to-train communicationare then presented, enhancing communication resource efficiency by reducing unnecessary data exchanges while preserving precious bandwidth resources. These mechanisms adapt to real-time network conditions, ensuring efficient platooning for CAVs and virtual coupling for railways. Additionally, resilient and secure control techniques designed to withstand, detect, and mitigate cyber threatsare discussed, safeguarding the integrity of networked transportation systems against various cyber-attacks such as denial-of-serviceand data falsification and replaying. Drawing on theoretical advancements, simulation results, and practical implications, this speech highlights how these advancements pave the way for a safer, more efficient, and resilient transportation ecosystem, addressing the pressing demands of future mobility.Finally, concluding remarks are drawn and some emerging challenges in the fieldare envisioned.
专家简介：Qinyuan Liu, Tongji University, China
报告题目：Industrial Visual Defect Detection for Intelligent Manufacturing
报告摘要：Driven by national “Intelligent Manufacturing”strategy, industrial machine vision has become a critical enabler for improving production yield and accelerating automation upgrades. Yet defects in real-world manufacturing environments are inherently high-dimensional, multifaceted, and hierarchical, which pushes defect detection technologies to evolve along a trajectory, from basic perception, through broad generalization, and ultimately toward deeper cognitive understanding. To meet this challenge, this report focuses on two pivotal manufacturing scenarios in industrial visual inspection: specialized line-specific performance and flexible cross-line generalization. We propose a three-pronged technical framework that integrates differential enhancement, diffusion-based distillation, and compositional learning. This approach is designed to empower intelligent manufacturing systems with defect detection that is both efficient and reliable, while also being adaptable to diverse production contexts.
报告专家：Zhenhua Wang, Harbin Institute of Technology, China
报告题目：Set-Membership State Estimation and Trustworthy Fault Detection for Uncertain Dynamic Systems
报告摘要：This report presents set-membership state estimation and trustworthy fault detection methods for uncertain dynamic systems. For set-membership estimation, it introduces new observer structures, simple yet efficient design strategies, novel parameter optimization approaches, and advanced set representation tools. Regarding fault detection, the report describes trustworthy fault detection methods based on set-membership estimation, along with application studies for satellite reaction wheels, underwater vehicles, and wastewater treatment processes.
报告专家：Minnan Luo, Xi’an Jiaotong University, China
报告题目：Multimodal Misinformation Detection with Limited Perception
报告摘要：Limited by privacy protection and platform restrictions, multimodal misinformation detection often suffers from limited perception, including incomplete information, insufficient data diversity, missing modalities, and scarce causal annotations. These challenges significantly undermine the robustness and generalization capability of existing detection models. This talk focuses on multimodal misinformation detection under limited perception in the era of foundation models, covering textual, image–text, and short-video misinformation detection. It will discuss several key research directions, including adversarial learning and knowledge transfer for continuously evolving generative misinformation, cross-modal understanding and fusion under limited semantic alignment, and information completion together with interpretable inference under missing modalities and insufficient causal supervision. The goal is to provide theoretical foundations and technical support for content security and intelligent governance in online social networks.
报告专家：Yan Liu, Northeast Forestry University, China
报告题目：Multimodal Brain-Computer Interfaces and Their Applications in the Diagnosis and Treatment of Brain Disorders
报告摘要：Multimodal brain-computer interfaces (BCIs), which integrate structural priors from brain imaging with functional electrophysiological signals, enable comprehensive analysis of brain structure and function. This approach holds great promise for precision diagnosis and treatment of brain disorders and represents a frontier direction in current brain science research. This report addresses this theme from three perspectives: research background, key scientific challenges, and the distinctive work carried out by our team. To tackle the core bottlenecks faced by BCIs--severe artifact contamination, the lack of a clear brain mechanism that constrains structure–function fusion, and the limited accuracy and dimensionality of current encoding/decoding techniques—our team has conducted systematic research on fast and robust artifact removal, high-precision spatial alignment between electrophysiology and structural imaging, accurate modeling of intracranial and extracranial electric field propagation, high spatiotemporal resolution source imaging, and adaptive personalized decoding algorithms. Some of these achievements have been applied to brain disorder related tasks, including epileptic spike detection, sleep seizure prediction, and auxiliary diagnosis of Parkinson’s disease. The report concludes with an outlook on future directions for leveraging multimodal BCIs to empower precision diagnosis and treatment of brain disorders.
报告专家：Weibo Liu, Brunel University of London, UK
报告题目：Towards Climate Resilience: Preliminary Studies on Extreme Wind Speed Prediction
报告摘要：Extreme wind speed events pose significant risks to infrastructure, renewable energy systems, and public safety, making reliable prediction an important component of climate resilience. However, AI-based extreme wind speed prediction faces two practical challenges: 1) how to learn robustly when training labels are noisy or imperfect; and 2) how to improve predictive reliability for medium-term forecasting and rare high-impact events. This talk presents two preliminary studies that address these challenges from complementary perspectives. First, to reduce the influence of noisy labels, a transfer-learning-assisted cooperative sample selection strategy is introduced, where multiple networks are jointly trained to identify potentially clean samples and improve model robustness. Second, a deep learning-based forecasting study is presented to compare representative time-series models under a unified ERA5-based evaluation pipeline, with attention paid to both point prediction accuracy and tail-event detection. Together, these studies provide an initial step towards climate-resilient extreme wind speed prediction under imperfect data conditions and medium-term forecasting requirements.
报告专家：Chaoqing Jia, Harbin University of Science and Technology, China
报告题目：Score-based Distributed Filtering Algorithm under Reputation-aware Mechanism
报告摘要：This talk is concerned with the distributed recursive filtering problem for a class of nonlinear dynamical systems over sensor networks with reputation-aware mechanism. A reputation-aware distributed recursive filtering algorithm is developed, where a novel reputation model is constructed to assign credibility scores to neighboring sensors, thereby identifying and rejecting unreliable data. Furthermore, a reputation-dependent recursive filter is designed, through which the upper bound of the covariance of filtering error dynamics is determined by solving a recursive matrix equation. The filter gain is subsequently parameterized by minimizing the trace of the derived upper bound so as to guarantee the optimality of estimation performance. Theoretical analysis establishes sufficient conditions for stability and performance of the proposed algorithm. To validate the feasibility and effectiveness of the developed approach, an illustrative case study is conducted.
科技处、自动化学院
2026年7月7日